Ferroelectrics such as lithium niobate (an oxide) possess a large non-resonant second-order optical nonlinearity which makes such materials useful for fabrication of a variety of optical and opto-electronic devices. Examples include optical switches and modulators, frequency shifting devices, polarized controllers, pulsed waveguide lasers, surface-acoustic-wave filters, and acousto-optic devices. These materials also often possess additional useful properties, such as piezoelectric, elasto-optic, and pyroelectric effects. Conventionally, such devices are fabricated from the bulk crystal material (typically a wafer about 0.5 to 1 mm thick), although most devices use only a small fraction of the surface volume of the material. Because these oxides tend to be chemically very inert, there are only a very limited number of surface modification tools (e.g., thermal diffusion) that can be used for fabrication purposes. It would be desirable if it were possible to deposit thin films of the ferroelectric materials on a substrate while controlling the composition and purity of the deposited materials. It would also be desirable if it were possible to deposit the film in a form which can be easily etched or ablated, permitting the fabrication of photolithograhically defined two-dimensional and three-dimensional structures on a planar substrate.
Numerous attempts have been made to grow crystalline LiNbO3 and other ABO3 ferroelectrics (where A and B are other metals) on various substrates. LiNbO3 thin films, for example, have been grown on semiconductors (e.g., Si and Ge), on dielectrics (e.g., MgO and Al2O3) and on ferroelectrics (e.g., LiTaO3 and LiNbO3 itself). In general, the objective of such deposition processes is to produce a crystalline thin film, since the crystalline form of the material usually has the best optical and electronic film qualities (e.g., optical transparency and nonlinear properties). Crystalline forms (e.g., single crystal textured, or polycrystalline) forms of these materials, however, etch very slowly with etchants current available. For example, a 50% aqueous solution of HF will have a negligible effect on single crystal LiNbO3, and reactive ion etching (RIE) using CCl2F2:Ar:O2 results in only about 3 μm/h etch rate. These etch rates are comparable to the etch rates for the masking materials that are used, making high resolution geometries essentially infeasible and resulting in very rough sidewalls with large optical losses. See J. L. Jackel, et al., “Reactive Ion Etching of LiNbO3, ” Applied Phys. Lett., Vol. 38, 1981, pp. 970 et seq.
Among the devices that utilize LiNbO3 are traveling wave electro-optic modulators. LiNbO3 traveling wave modulators are currently formed utilizing a LiNbO3 substrate containing a Mach-Zehnder waveguide geometry, a buffer layer (a thin dielectric film such as SiO2 isolating the light in the waveguide from the metal electrodes), and metal electrodes in the form of a microwave strip line. State of the art commercial traveling wave modulators (TWMs) using these structures have a 7 GHz bandwidth (corresponding to 10 Gb/s maximum transmission rate for non-return to zero (NRZ) coding) and an operating voltage at the maximum speed of Vπ@7 GHz=6 volts. At 40 Gb/s (30 GHz bandwidth, NRZ), numerical simulation shows that the conventional LiNbO3 TWM requires a drive voltage of about 9 volts with an electrode length L=1.6 cm and thickness te=30 μm. However, the available gallium arsenide drive electronics at this bit rate has a maximum voltage swing of about ±4.5 volts. Thus, the conventional TWM structure would theoretically be capable of attaining the 40 Gb/s bit rate, but there is no margin of error to allow for processing variability. To account for thermal voltage degradation and process variations in the electronics, a margin of error of about 10% must be allowed (i.e., the TWM must be capable of operating at ±4 volts).
Noguchi, et al. (“A Broadband Ti:LiNbO3 Optical Modulator with a Ridge Structure,” J. of Lightwave Technology, Vol. 13, No. 6, June 1995, pp. 1164-1168) have shown that etching 3-4 μm deep ridges in the LiNbO3 above the Mach-Zehnder waveguides produces a better overlap between the optical and microwave fields, thereby allowing the drive voltage to be reduced. However, difficulties are encountered in making commercial devices having such structures because, as noted above, the etch rates of crystalline LiNbO3 are very slow. The resulting surfaces are rough, significantly increasing the waveguide's propagation loss. In addition, the reliability of devices made using present etching techniques is questionable. A variation of this approach is shown in U.S. Pat. No. 6,172,791 to Gill, et al., in which ion implantation is used to allow etching at an angle to form ridges with reentrant sidewalls to further shape the electric field in the ridges.